Website CV Paper
Fields
Behavioral & Experimental
Job market paper
Designing Experiments to Distinguish Theories
We develop a measure of an experimental design's ability to distinguish between competing parametric theories. A design distinguishes a theory from its competitors when these competitors are unable to reproduce behavior predicted by the theory. We propose an algorithmic procedure for constructing maximally-distinguishing experimental designs, given a domain and hypothesized parameter values. Using this procedure, we generate 60 algorithmic designs to distinguish between competing theories of choice under risk. We show experimentally that more distinguishing designs generate stronger statistical results; and that experimental results are more likely to favor a given theory when distinguishability is asymmetric, i.e. high for the theory, low for its competitor. We then use distinguishability to analyze a novel dataset of 123 lottery choice experiments from the experimental risk literature. We provide suggestive evidence that algorithmic designs substantially improve upon the state of the art for design development and demonstrate how distinguishability may change the interpretation of canonical experimental results.